How tameyer41/liftoff works
Project Type & Purpose
• This is a full-stack web application (Next.js) that simulates mock interviews and delivers AI-powered feedback.
• The user interacts via their browser (React UI), records a video, which is transcoded client-side, then sent to serverless API routes that call OpenAI’s Whisper (for transcription) and GPT-3.5 Turbo (for feedback). Rate limiting uses Upstash Redis.
Main Components
A. Client (Browser/Frontend)
• React pages & components (app/page.tsx, pages/demo.tsx, components/Gradient.js)
• FFmpeg.wasm (public/ffmpeg/dist/ffmpeg-core.*) used in-browser to transcode raw webcam video to MP3.
B. Next.js Serverless Layer (Vercel Edge/Node)
• API Routes under pages/api:
• transcribe.ts → calls OpenAI Whisper
• generate.ts → streams chat responses via OpenAIStream (GPT-3.5-Turbo)
• blocked.ts → checks rate limits via Upstash Redis
• Middleware (middleware.ts) intercepts requests for authentication/rate-limiting.
• app/opengraph-image.tsx runs at the edge for dynamic OG image generation.
C. External Services
• OpenAI API (Whisper & GPT-3.5 Turbo)
• Upstash Redis (serverless Redis for rate limiting)
• Vercel (hosting, edge network, CDN for static assets & edge functions)
Data Flow & Interactions
1. User opens the app → Next.js SSR/SSG renders pages. Static assets served from Vercel CDN.
2. User records video via React‐Webcam. FFmpeg.wasm transcodes on client.
3. Frontend POSTs MP3 data to /api/transcribe
↳ Edge function invokes OpenAI Whisper → returns transcript.
4. Frontend streams transcript to /api/generate
↳ Edge function uses OpenAIStream → returns streaming AI feedback.
5. Each API call passes through middleware.ts → blocked.ts uses Upstash Redis to implement rate limits.
Architectural Patterns & Design Principles
• Separation of Concerns: UI vs. API vs. external services.
• Serverless functions as micro-endpoints for transcription, generation, and rate-limiting.
• Client-side heavy lifting (FFmpeg) to minimize server load & simplify streaming.
• Edge rendering for performance (SSR, dynamic OG images).
Final Checklist
[ ] All major components (Browser, Next.js, API routes, Redis, OpenAI) present
[ ] Notes on architectural patterns (serverless, edge rendering, client transcoding)